I’ve been building Orthodox Theology, a specialized ChatGPT plugin for people interested in Orthodox theology, Scripture, patristics, Church history, Councils, Liturgy, and primary-source research.
What I wanted to solve was a problem I kept seeing with general-purpose AI:
Finding relevant sources is not the same thing as knowing what those sources actually prove.
For example:
- One Church Father teaching something does not automatically mean “Orthodoxy teaches this.”
- Several quotations do not automatically establish consensus.
- A quotation repeated across websites is not necessarily authenticated.
- A translation should not silently include an interpretation.
- Failing to locate a source does not prove that the source does not exist.
So instead of just prompting ChatGPT to “answer from an Orthodox perspective,” I built explicit reasoning and verification rules around it.
What it does
Orthodox theological reasoning
It treats Scripture, the Fathers, Councils, and Liturgy together within Holy Tradition and distinguishes the different roles those sources play.
Doctrinal-status analysis
It distinguishes between things such as:
- dogmatically defined teaching;
- broadly received Orthodox teaching;
- admissible but nonbinding theological opinions;
- genuinely unresolved questions;
- historical opinions whose wider status is not established;
- claims where the available evidence is insufficient.
Patristic source verification
It is designed to be strict about:
- exact quotations;
- attribution;
- editions;
- translations;
- page and section numbers;
- surrounding context.
A search snippet or commonly repeated citation is not automatically treated as direct verification.
Scripture interpretation
It keeps separate:
- quotation;
- literal translation;
- simplified translation;
- interpretive paraphrase;
- exegesis.
It also preserves differences between textual witnesses such as the Septuagint/LXX, Masoretic Text, Greek, Hebrew, and named translations instead of silently harmonizing them.
Source hierarchy
A Council, broad ecclesial reception, one Father, a local jurisdiction, and a modern academic scholar are not treated as five equivalent citations.
The system asks what each source is actually evidence for.
Why I think this matters beyond theology
Building this made me realize that a major challenge in domain-specific AI is not just retrieval.
It is epistemic control:
«What kind of evidence is this?
What claim does it support?
How strongly does it support it?
What conclusions are not justified by it?»
That seems applicable to law, medicine, history, policy, philosophy, scientific literature, and any field where sources have different levels of authority or reliability.
I’m looking for people to test it
If you want to try to break it, give it difficult cases such as:
- disputed patristic quotations;
- claims of Orthodox consensus;
- competing interpretations among the Fathers;
- alleged theologoumena;
- LXX vs. Masoretic differences;
- suspicious source citations;
- “Is this actually Orthodox teaching?” questions;
- difficult biblical passages.
I’m especially interested in failures.
If it invents consensus, mishandles a source, confuses interpretation with translation, or makes a claim stronger than the evidence supports, I want to know about it.
The plugin is called Orthodox Theology and is available in ChatGPT.
If you work on domain-specific AI systems, I’d also be interested in how you handle the difference between retrieving evidence and deciding what that evidence is actually allowed to establish.
